AI Chatbot Integration Services

Description

We run a customer-facing web app and a Zendesk helpdesk, and we want a chatbot that sits across both rather than a widget bolted onto the marketing site. Roughly 60% of our inbound tickets are the same eight questions about order status, refunds and delivery windows, and our support team spends its day copying answers between tabs. The work is integration rather than model training. You will connect an existing LLM provider to our order API and knowledge base, decide what the bot may answer on its own and what it must hand to a human, and build the handover so the agent picking it up can see the whole conversation. We care more about the escalation path being right than about the bot answering everything. You should be comfortable saying "the bot should not attempt this" — a confident wrong answer about a refund costs us more than a handover.

• Connect the chatbot to our order API so it can answer status questions with live data, not a scraped FAQ
• Define the escalation rules — which intents go straight to a human, and what the bot must never answer alone
• Build the Zendesk handover so an agent inherits the full conversation transcript and any data the bot fetched
• Add conversation logging we can review weekly to find questions the bot got wrong
• Write a fallback path for when the LLM provider is slow or down, so the widget degrades to a contact form

• Chatbot integrated into the web app and Zendesk, deployed to staging then production
• Escalation rule set, documented and editable by our support lead without a developer
• Conversation log dashboard with weekly export
• Runbook covering provider outage, rate limits and rollback